Ready to go
in minutes

1. Prerequisites

.NET 10 SDK as well as Ollama — runs locally, no cloud connection required. After installation, pull the appropriate models:

CPU mode (default, smaller model, runs without a GPU):

• ollama pull phi3.5:3.8b-mini-instruct-q4_0
• ollama pull all-minilm

GPU mode (larger model, better answer quality, requires a compatible GPU on the Ollama host):

• ollama pull llama3.1:8b-instruct-q4_K_M
• ollama pull bge-m3

2. Build & first start

In the project directory:

• dotnet restore SupportFlowAI.slnx
• dotnet build SupportFlowAI.slnx
• dotnet run --project src/SupportFlowAI.Api

On first start, the app automatically creates:

• data/supportflowai.db — SQLite database, created via EF Core migration
• data/dataprotection-keys/ — key ring for encrypted credentials & auth cookies

Then open the address in your browser (see the console output, typically https://localhost:56657 or http://localhost:56658 in dev mode).

3. First configuration highlight: vector index mode

In src/SupportFlowAI.Api/appsettings.json, VectorIndex:Mode controls the RAM/speed trade-off of the knowledge search — a change requires an application restart:

4. Configuration files at a glance

src/SupportFlowAI.Api/appsettings.json

• Database:ConnectionString — path to the SQLite database
• VectorIndex:Mode — see above, restart required

data/model-settings.json (created automatically by the initial setup)

• Ollama:ChatModel / Ollama:EmbeddingModel — takes effect immediately, no restart needed

data/mailsync-settings.json (written via /Admin/MailSync)

• MailSync:Enabled — on/off switch for automatic background synchronization
• MailSync:IntervalMinutes — interval between two automatic runs, takes effect immediately

data/supportflowai.db

• The actual SQLite database. For a complete reset: stop the application, delete this file (and if applicable data/dataprotection-keys/), then restart — the initial setup will appear again.